The driving question behind this project is “WHERE are energy efficient homes in the Bay Area and WHO benefits from occupying them?” Behind this question is a foundational understanding that occupying an energy efficient home has financial benefits. The most obvious advantage is that utility bills are lower than they otherwise would be, resulting in cost savings that accumulate over time. It has also been documented that energy efficient homes earn a greater resale value (LINK?). Achieving energy efficiency, however, costs money, either through retrofits of an old home or brand new construction (which is increasingly rare in the highly developed Bay Area). Due to this barrier to entry, it would logically follow that residential energy efficiency is one of the countless mechanisms in our society that perpetuates economic disparity. The following analysis seeks to view this issue from a number of angles, picking it apart with statistical analyses and placing on the map what the energy consumption landscape looks like in the Bay Area.
Energy consumption data comes from aggregated PG&E data, via another data project called “The Bay Area Energy Atlas” from the organization BayREN (LINK?). The data file I used included a combined annual electricity and natural gas usage figure for each year from 2013 to 2017, aggregated to the Census Tract level but split into categories of building type (residential, industrial, etc.). For my analysis, I primarily used the residential sub-category: single-family homes. In addition to the combined usage figure, numbers on population and ‘built square feet’ for each building type allowed for calculation of varying consumption metrics (detailed below).
Socio-demographic and housing characteristics data for cross-analysis was mostly brought in from ACS 5-year estimate surveys, which allowed for selection of individual years in order to match the time frame of the PG&E data. In the case of race (used for equity analysis) data came from the 2010 census.
(Note: earlier on in the project, an attempt was made to incorporate all five years of data and build dashboards with year-selection menus, so that maps and plots could be created for each year. In the end, however, due to time constraints, analysis was only performed for the year 2013).
Energy Intensity is a standard measure of building energy efficiency which is typically expressed in Btu/SF or kBTu/SF. Because it is normalized by space size, it allows for comparison between buildings of varying sizes. Below is a map showing averaged EISF values for Bay Area tracts. A darker orange means higher energy intensity and on average less efficient buildings, and vice versa.
Instead of measuring energy consumption at the building level (for which Energy Intensity is the default measure), considering per person usage could offer unique insights. The PG&E data gathered through BayREN contained a ‘Usage Per Capita’ column for all residential buildings, allowing for the derivation of the ECPC measure. (Note: ECPC was derived using the proportionality of single-family homes to total residential buildings by built square feet. ACS and Census data was searched at length for a population figure of people living in ‘single-family homes’ to no avail). The map below shows varying levels of ECPC for Census tracts across the Bay Area.
For this metric, ACS data was collected (from table DP04: “Selected Housing Characteristics”) that provided the number of “1-unit” households in each tract. For purposes of this analysis, “1-unit” households are equated to “single-family” homes. Thus, the single-family energy consumption number from the PG&E data was divided over the ACS data on number of 1-unit households per tract to arrive at this metric for household consumption, averaged across each tract. ECPH is similarly mapped below for 2013.
Having established the three metrics for energy consumption and efficiency above (EISF, ECPC, and ECPH), this data was cross-examined with selected socio-demographic and housing characteristics which might have links with energy use. Simple linear regression was used throughout as a useful method for assessing the strength of these possible relationships. On the social side, the two variables examined below are income and race. Takeaways from these analyses are relevant to discussions of equity, with the understanding described above that energy efficiency perpetuates financial disparities between groups. On the housing side, three variables were assessed: tenure statues (whether an occupant owns or rents), the number of years spent living in a home, and the value of a home. These variables (five in total), came from ACS and Census data, and with the exception of race, match the year of the energy data used (2013). As a reminder, data was necessarily collected at the tract level to match the smallest possible geographic granularity offered for energy consumption.
A weighted average income figure was calculated for each tract by multiplying the number of people in each income bracket by the midpoint income of that bracket.
We see in the plot above that, as suspected, income has a negative association with energy intensity. In other words, as income goes up, people’s homes tend to get more energy efficient. House size is also incorporated here and displayed by color, with tracts having on average larger homes showing up in lighter shades of blue. Also as expected, house size increases with increased income.
There seems to be a low cluster of more efficient tracts separated from the pack..
Where are those more efficient tracts?
Noticeably, big urban centers are missing from this map (San Francisco, Oakland, San Jose). The map seems to be made up of less populated tracts with more open space, and towns on the Peninsula (San Mateo, South San Francisco, etc.). The lower EISF numbers in the more “countryside” tracts could be a result of homes that are not regularly used and thus sit vacant much of the time. Also note, for sparsely populated tracts where the base of data observations is much smaller, efficiency averages can be more easily influenced by a smaller number of buildings. The low EISF tracts on the Peninsula, in the corridor leading to Silicon Valley could be a product of tech-oriented residents who work in the technology industry and likely have the means and education to invest in home efficiency measures. These are just theories for many possible reasons leading to this cluster.
When we change our consumption metric from Energy Intensity to Per Capita consumption (ECPC), we see that on an individual level, people tend to consume more as their income goes up. There is an interesting relationship here, because we saw above that as income goes up, HOUSES tend to be more efficient (per SF). The confounding variable leading to the opposite slopes of these two graphs is almost certainly house size, in that EISF numbers will shrink as income rises and houses become more spacious (as the denomenator of square feet increases). This, in combination with the fact that efficient homes (which cost more money) will be occupied by higher-income inhabitants, allows ECPC to rise as income rises while EISF falls.
Equity analysis in this case is tricky, since consumption data from the utility (PG&E) was, for privacy reasons, aggregated to the census tract level as the smallest geographical granularity. This means that we can compare a few census tracts by their overall efficiency scores (EISF for single family homes) and their overall racial makeup, but we cannot say for certain which specific houses operate more or less efficiently and who exactly lives in those houses. So, the tracts with both the highest and the lowest EISF scores (shown in the EISF maps) are inspected for their racial make up to see if any trends are apparent.
As we can see above, the tract with the most efficient homes (located in a hilly and coastal region of San Mateo county) is much less diverse than the tract with the highest EISF (located in Pittsburg). The sample size here is small, but the implications for energy efficiency and equity speak for themselves.
##
## Call:
## lm(formula = ecph ~ owned_units_perc, data = ECPH_vs_TENURE_2013)
##
## Residuals:
## Min 1Q Median 3Q Max
## -39261390 -7242051 -1431296 4866181 202095189
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 28578202 1101564 25.94 <2e-16 ***
## owned_units_perc 56948772 1733185 32.86 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 13850000 on 1367 degrees of freedom
## Multiple R-squared: 0.4413, Adjusted R-squared: 0.4409
## F-statistic: 1080 on 1 and 1367 DF, p-value: < 2.2e-16
##
## Call:
## lm(formula = ecpc ~ owned_occ_perc, data = ECPC_vs_TENURE_2013)
##
## Residuals:
## Min 1Q Median 3Q Max
## -20360763 -3875489 -271107 2890330 64527492
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1996103 517724 -3.856 0.000121 ***
## owned_occ_perc 32557036 808282 40.279 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 6278000 on 1364 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.5433, Adjusted R-squared: 0.5429
## F-statistic: 1622 on 1 and 1364 DF, p-value: < 2.2e-16
The above (2) plots show that increased ownership is associated with increased consumption, both when measured per household and per occupant
##
## Call:
## lm(formula = energy_intensity ~ owned_units_perc, data = EISF_vs_TENURE_2013)
##
## Residuals:
## Min 1Q Median 3Q Max
## -44493 -4502 1289 5325 54132
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 51485.6 914.1 56.33 <2e-16 ***
## owned_units_perc -14675.9 1436.1 -10.22 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 11480 on 1382 degrees of freedom
## Multiple R-squared: 0.07026, Adjusted R-squared: 0.06959
## F-statistic: 104.4 on 1 and 1382 DF, p-value: < 2.2e-16
##
## Call:
## lm(formula = ecph ~ weighted_avg_years_in_home, data = ECPH_vs_YEARS_in_home_2013)
##
## Residuals:
## Min 1Q Median 3Q Max
## -37873158 -8410277 -2560795 4795491 204376242
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 9797196 2399111 4.084 4.69e-05 ***
## weighted_avg_years_in_home 2568290 114773 22.377 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 15850000 on 1367 degrees of freedom
## Multiple R-squared: 0.2681, Adjusted R-squared: 0.2676
## F-statistic: 500.7 on 1 and 1367 DF, p-value: < 2.2e-16
##
## Call:
## lm(formula = ecph ~ weighted_avg_value, data = ECPH_vs_VALUE_2013)
##
## Residuals:
## Min 1Q Median 3Q Max
## -50599004 -9983583 -115997 8491055 199325888
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.219e+07 1.135e+06 37.16 <2e-16 ***
## weighted_avg_value 3.463e+01 1.771e+00 19.56 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 16360000 on 1366 degrees of freedom
## (1 observation deleted due to missingness)
## Multiple R-squared: 0.2188, Adjusted R-squared: 0.2182
## F-statistic: 382.6 on 1 and 1366 DF, p-value: < 2.2e-16
In conclusion, the main driving hypothesis of this analysis seems to be confirmed: that energy efficient homes tend to be owned and occupied by more members of higher income brackets. An interesting related trend was also revealed: that higher earners also consume more on a per capita basis. So by one view (using EISF), we might applaud the efforts of wealthy homeowners who make their homes more energy efficient. At the same time, however, wealthier indviduals appear to consume more per capita (measured by ECPC). This consumption appears to be masked by the efficiency of the homes in question, as well as the size of the homes in question (as a larger but equally consuming building will have a lower EISF). Our equity analysis also shed some light on the racial makeup of the most extreme tracts in the analysis (minimum and maximum EISF); the tract with the highest efficiency was predominantly White with some Asian population, while the tract with the lowest EISF scores was much more racially diverse. Both conclusions point to some of the disparities surrounding residential energy efficiency. And while there should certainly be a goal to ‘democratize’ building energy efficiency for environmental reasons, we should also examine our apparent tendency to consume more energy as we earn more and can afford more efficient homes.